housing construction in stockholmkth.diva-portal.org/smash/get/diva2:1194191/fulltext01.pdf ·...
TRANSCRIPT
DEGREE PROJECT REAL ESTATE AND CONSTRUCTION MANAGEMENT BUILDING AND REAL ESTATE ECONOMICS MASTER OF SCIENCE, 30 CREDITS, SECOND LEVEL STOCKHOLM, SWEDEN 2018
Housing Construction in Stockholm
Fundamental Factors’ Impact on Construction Volume
Lovisa Onsbring
Master of Science thesis
Title Housing Construction in Stockholm - Fundamental Factors’ Impact on Construction Volume
Author Lovisa Onsbring Department Department of Real Estate and
Construction Management Archive number TRITA-ABE-MBT-184 Supervisor Sviatlana Engerstam
Keywords Housing construction, Swedish real estate market, Housing market, Fundamental economic factors
Abstract
Housing availability plays a central role for the mobility in the labour market and for the economic
growth in a country. Housing construction in Sweden is now increasing strongly after a period of
low construction during and after the financial crisis in 2008. Despite this, many of the Swedish
municipalities indicate that there is a housing shortage, especially in the metropolitan regions.
The main reason for this is that the population has grown at a faster pace compared to the number
of homes.
This thesis uses a regression analysis to analyse what fundamental economic factors that affected
housing construction in Stockholm County between 1992 and 2016.
The result from the regression analysis shows that the repo rate, inflation, GDP, population,
unemployment, housing prices, and construction cost, with a lag of seven, seven, zero, one, six,
two, and three quarters respectively, have a significant effect on housing construction.
Furthermore, the regression model showed a low significance level for income, thus this variable
was excluded from the model. Moreover, a time trend was included in the regression model. The
time trend shows statistically significance, and its β coefficient implies an approximate 3%
increase in housing construction per year, on average.
Moreover, the regression model, presented in this thesis, has an R-squared value of 0.73, which
indicates that the model can explain 73% of the variation in housing construction in Stockholm
County between 1992 and 2016.
Examensarbete
Titel Bostadsbyggande i Stockholm – Fundamentala faktorers inverkan på konstruktionsvolym
Författare Lovisa Onsbring Institution Fastigheter och byggande Arkiv nummer TRITA-ABE-MBT-184 Handledare Sviatlana Engerstam Nyckelord bostadsbyggande, svenska
fastighetsmarknaden, bostadsmarknaden, fundamentala ekonomiska faktorer
Sammanfattning
Tillgängligheten av bostäder spelar en central roll för rörligheten på arbetsmarknaden och för den
ekonomiska tillväxten i ett land. Bostadsbyggandet i Sverige ökar nu kraftigt efter en period av
lågkonjunktur, under och efter finanskrisen 2008. Trots detta anser många kommuner i Sverige att
bostadsbristen är stor, särskilt i storstadsregionerna. Den främsta orsaken till detta är att
befolkningen har vuxit i en högre takt än antalet bostäder.
Denna uppsats använder en regressionsanalys för att analysera vilka fundamentala ekonomiska
faktorer som påverkat bostadsbyggandet i Stockholms län mellan 1992 och 2016.
Resultatet av regressionsanalysen visar att reporäntan, inflationen, BNP, befolkningsmängd,
arbetslöshet, bostadspriser och byggkostnader, med sju, sju, noll, en, sex, två, respektive tre
kvartals förskjutning, hade en signifikant inverkan på bostadsbyggandet. Vidare visade
regressionsmodellen en låg signifikansnivå för variabeln inkomst, vilket medförde att denna
variabel uteslöts från modellen. Dessutom inkluderades en tidsvariabel i regressionsmodellen.
Tidsvariabeln visar statistisk signifikans och dess koefficient indikerar att bostadsbyggandet i
genomsnitt ökade med ca 3% per år.
Regressionsmodellen som presenteras i denna uppsats har ett R2-värde på 0,73 vilket indikerar att
modellen förklarar 73% av variationen i bostadsbyggande i Stockholms län mellan 1992 och 2016.
Acknowledgements
First, I would like to thank my supervisor Sviatlana Engerstam for her valuable guidance,
enthusiasm and constructive feedback throughout the whole process of writing this thesis.
Furthermore, I would also like to express my great gratitude to Han-Suck Song, at the department
of Real Estate and Construction Management at KTH, for all the valuable discussions, and for
sharing his knowledge, as well as giving his input on the econometrics of this study.
Table of contents
1 Introduction....................................................................................................................................1
1.1 Background............................................................................................................................1
1.2 Purpose..................................................................................................................................3
1.3 Researchquestion..................................................................................................................4
1.4 Delimitations..........................................................................................................................4
2 OverviewoftheSwedishhousingmarket.......................................................................................5
2.1 Theroleofpolicyregulations................................................................................................5
2.1.1 Governmentregulations....................................................................................................5
2.1.2 Theplanningprocess.........................................................................................................5
2.2 Stockholmhousingmarket....................................................................................................7
3 Literaturereview.............................................................................................................................9
3.1 Businessandrealestatecycles..............................................................................................9
3.2 Behaviourbasedtheories....................................................................................................13
3.3 Theqtheoryofhousinginvestment....................................................................................13
3.4 Housingbubbles...................................................................................................................14
3.5 TheDiPasquale-Wheatonmodel.........................................................................................15
4 Methodologyandresearchdesign................................................................................................17
4.1 Literaturereviewasatheoreticalbasisfortheresearch....................................................17
4.2 Quantitativemethod...........................................................................................................18
4.3 Regressionanalysis..............................................................................................................18
4.3.1 Linearregressionmodels.................................................................................................18
4.3.2 Distributedlagmodels.....................................................................................................19
4.3.3 Timetrend.......................................................................................................................19
4.4 Statisticaltestsofthedataandmodels...............................................................................20
4.4.1 Stationarity......................................................................................................................20
4.4.2 Correlation.......................................................................................................................20
4.4.3 Multicollinearity..............................................................................................................20
4.4.4 Autocorrelation...............................................................................................................21
4.4.5 Heteroscedasticity...........................................................................................................21
5 Datacollectionandhypotheses....................................................................................................22
5.1 Variables..............................................................................................................................23
5.1.1 Thedependentvariable...................................................................................................23
5.1.2 Independentvariablesandhypotheses...........................................................................23
6 Results...........................................................................................................................................28
6.1 Regressionoutcomes...........................................................................................................28
6.2 Hypothesistesting...............................................................................................................31
7 Analysis..........................................................................................................................................32
7.1 Analysisofvariables.............................................................................................................32
7.2 Limitations...........................................................................................................................35
8 Conclusion.....................................................................................................................................37
8.1 Concludingremarks.............................................................................................................37
8.2 Suggestionsforfurtherresearch.........................................................................................37
9 References.....................................................................................................................................38
Appendices............................................................................................................................................41
AppendixI.Regressionmodel...........................................................................................................41
AppendixII.Statisticaltesting...........................................................................................................42
Test for Stationarity.......................................................................................................................42
Test for Correlation.......................................................................................................................44
1
1 Introduction
This chapter serves as an introduction to the research subject. The chapter presents the
background of the thesis together with the purpose, research question and delimitations of the
thesis.
1.1 Background
There are about 4.8 million apartments in Sweden. In a short-term perspective, the number of
apartments available in Sweden is constant, and due to the immobility of buildings, these apartments
are not always in the best geographical place for optimal utilization. Whether there is a housing
shortage or not in Sweden, depends on how we utilize the existing stock, as well as the rate that we
build new and demolish existing apartments. (Boverket, 2011)
Due to increasing income, low interest rate, low amortization rate and a high population growth,
the demand for housing has increased significantly in recent years. (Boverket, 2014a) Over the last
two years, there has also been a significant increase in housing construction. According to the
Swedish National Board of Housing, Building and Planning (Boverket), new constructions of
approximately 67,000 homes begun in 2016. This is an increase of more than 30 percent compared
to 2015. (Boverket, 2017b)
Housing availability is important in many aspects. It is necessary for the labour market to work and
for the economic growth of society to continue. The future need for housing, and where this need
is located, is therefore of great importance to the economy. Housing investments based on short-
term market conditions can be inaccurate in a longer perspective due to the long life and the
geographical immobility of buildings. It is therefore important to make long-term estimations of
housing needs. (Boverket, 2015)
Moreover, the rapid urbanization in Sweden after World War II resulted in an urgent need for
housing in the cities. This lead to the so-called Million Programme, initiated by the Swedish
government. The aim of the program was to build one million homes during a decade, to overcome
the current housing shortage, and to increase the standard of living. In the years 1965-1974, about
1,006,000 homes were built in Sweden. After the program was completed, housing construction
2
slowed down during a ten-year period, after which it began to rise again in the mid-1980s.
(Boverket, 2014b) Figure 1 illustrates the housing development in Sweden between 1950 and 2016.
Furthermore, the development in the housing and real estate market has through history played a
leading role in economic crises, where the Swedish crisis in the early 1990s, and the subprime
mortgage crisis in the U.S. between 2007-2010 are two examples. Heavy drops in real estate prices
have in many countries been associated with major disruption in the country’s economy. Usually
these price drops have been preceded by a long period of rising house prices and, in many cases, an
increase in the indebtedness in the household sector.
However, the housing prices in Sweden did not fall to the same extent as they did in many other
countries during the financial crisis in 2008. One potential reason to this is that housing supply has
been low in relation to demand, partly due to the low housing construction in Sweden since the
early 1990s.
In the early 1990s, major changes to tax policy were made, which had a negative effect on housing
construction. The combined effect of changes in the VAT system, raised property tax, reduced
subsidies, sharply increased real interest rates after tax and a deep recession significantly reduced
the demand for new housing. (Sveriges Byggindustrier, 2015)
Figure 1. Housing construction in Sweden 1950-2016. (Sveriges Byggindustrier, 2016)
0
20 000
40 000
60 000
80 000
100 000
120 000
1950
1952
1954
1956
1958
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
2012
2014
2016
One- andtwo-familyhousesMulti-dwellingbuildings
Quantity
Year
3
Moreover, in a housing market survey from 2017, conducted by the Swedish National Board of
Housing, Building and Planning, many municipalities indicated an imbalance in the local housing
market. 255 out of a total of 290 municipalities in Sweden estimated that there is a deficit of housing
in their municipality. This is an increase in housing deficit compared to the year before. According
to the survey, only one municipality estimates an excess in housing. The remaining 34
municipalities considered the housing market to be in balance.
In the survey, the Swedish municipalities also rated what they consider to be the three main barriers
to increase housing construction. Results from the survey showed that high production cost was
seen as the main barrier. Also, lack of detail plan on attractive land and difficulties for individuals
to get loans, was pointed out as limiting factors for housing construction. (Boverket, 2017a)
1.2 Purpose
The overall purpose of this thesis is to analyse what fundamental economic factors that affect
housing construction in Stockholm. There are many different factors that might affect housing
construction. A more market driven demand for housing in Sweden can be seen after 1992, which
indicates that housing construction might be explained by fundamentals, to a larger extent, after
1992. However, the planning system and some other policy regulations in Sweden still influence
the development on the housing market. Municipalities have a strong control over housing
construction, and it is therefore of great importance to evaluate if municipalities are in line with
economic development and housing demand. Housing shortage is a huge problem in some regions
in Sweden, especially in Stockholm. To reduce problems with housing shortage, the underlying
causes for housing construction must be investigated.
This thesis provides deeper insight into how fundamental economic factors affect the level of
housing construction in Stockholm. This is the first study of this kind and it contributes to the
research gap that is found in the literature review. In this thesis, fundamental economic factors are
factors that are in line with economic theory and that are valid in several previously performed
empirical studies.
The results presented in the thesis might be used by municipalities to achieve a better understanding
of the impact of economic variables on housing supply, which could help them to improve their
planning policies.
4
1.3 Research question
The thesis aims to answer the following research question:
- Can housing construction in Stockholm be explained by fundamental economic factors, and
if, to what extent?
1.4 Delimitations
The thesis is delimited to analyse fundamental factors that are quantifiable, meaning that qualitative
factors are excluded from the analysis in this thesis. However, political factors, such as
governmental reforms, and the structure of the Swedish planning system, are briefly described in
the following chapter, together with relevant literature on the subject. However, the political factors
are not included in the regression analysis, although they may have a major impact on housing
construction in Stockholm.
5
2 Overview of the Swedish housing market
This chapter presents some of the qualitative factors that may affect housing construction in
Sweden. Political influences, such as governmental reforms, and the structure of the Swedish
planning system, are described together with relevant literature. Furthermore, the chapter also
provides a description of the housing market in Stockholm; its development over the last decades,
and the current market situation.
2.1 The role of policy regulations
2.1.1 Governmentregulations
Lind (2003) argues that the main reason for the large variation in housing construction, seen in
Sweden over time, is the Swedish housing policy. This involves, for example, changes in taxes, and
the introduction of government subsidies for housing construction. (Lind, 2003)
The large decline in housing construction in the beginning of the 1990s was partly due to the
financial crisis in Sweden and the recession that followed. However, also the major changes in
economic policy that resulted in a more market driven demand for housing investment played an
important role in the decline in housing construction. Changes in connection to the tax reform in
1993 affected the housing market. A decline in interest subsidies and interest rate guarantees for
construction projects meant that the government-subsidised secondary mortgages disappeared,
leading to a market with similar funding conditions for both public and private housing companies.
(Emanuelsson, 2015)
2.1.2 Theplanningprocess
In Sweden, municipalities have a monopoly on spatial planning, meaning that the municipality
decides how the land will be used and to what extent it will be built within the municipality. Because
of the municipal planning monopoly, the municipality serves as a key actor for housing
development.
Furthermore, the planning process in Sweden is regulated in the Planning and Building Act (PBL).
All municipalities must have a comprehensive plan that provides guidance on how the land and
water areas, and the existing urban environment should be used, preserved and development in the
6
long term. The comprehensive plan should cover the entire municipality. However, this plan is not
legally binding. (Boverket, 2016)
The detailed development plan, on the other hand, is legally binding, and it serves as the
implementation instrument for the municipality. This plan can only be produced and adopted by the
municipality itself, and it regulates how land and water within an area in the municipality should be
used and built. For example, requirements for the execution of buildings can be set in the plan, with
details such as how large or tall a building may be, the materials to be used, or the distance between
buildings. This allows the municipality to control all housing development taking place within the
municipality.
The implementation period for a detailed development plan must be set to a minimum of five years
and a maximum of 15 years, meaning that the rights that the plan accord are economically protected
during this period. (Boverket, 2014c)
The planning process in Sweden has many times been pointed out as complicated and time
consuming. There are many requirements that need to be fulfilled in order to adopt a new detailed
development plan. However, a change in law in 2015 aimed to simplify the planning process in
Sweden. What was previously called “simple planning procedure” is now the norm and an extended
planning procedure applies under certain circumstances, for example, when the detailed
development plan is not compatible with the comprehensive plan, or when the plan can be assumed
to cause a significant environmental impact.
For the municipality to adopt a detailed development plan through a standard procedure, the
municipality must first draw up a planning proposal, then consulted about the proposal, and finally
had it reviewed. The time required for each step in this process varies, which make the planning
process unpredictably. (SFS 2010:900)
Furthermore, Mayer and Somerville (2000a) examined the relationship between land use regulation
and housing construction. By using quarterly data from 44 U.S. metropolitan areas between 1985
and 1996, they found that cities with more extensive regulations have a lower level of new
construction. The mean construction start was found to be up to 45 percent lower in these cities.
Further, their study also shows that metropolitan areas with greater land use regulations experience
lower price elasticities compared to cities with less regulations. However, Mayer and Somerville
divided regulation into three different categories, depending on whether they add explicit costs,
uncertainty, or delays to the development process. Their results, in accordance with several other
studies, showed that regulations that lengthen or causing uncertainty to the development process
7
have larger and more significant effects on new construction than development fees and other
regulations that add costs to the development process. (Mayer and Somerville, 2000a)
In addition, several studies have found a correlation between the complexity of the planning
process, the level of new construction, and the prices of new constructions. Luger and Temkin
(2000) compared states in the U.S., and found that strict regulation, low level of housing
construction, and high housing prices correlated. Furthermore, Glaeser and Gyourko (2002) and
Boulhouwer and deVries (2002) found a similar relationship for the U.S. and Holland, respectively.
2.2 Stockholm housing market
Stockholm County has a population of almost 2.3 million people, spread over approximately one
million homes. About 40 percent of the Swedish housing stock consists of tenant owned apartments
(bostadsrätter), 36 percent of rental apartments (hyresrätter), and 24 percent are ownership
apartments (äganderätter). (Länsstyrelsen Stockholm, 2017)
Stockholm is one of the fastest growing regions in Europe, with a population growth of over 320,000
people during the last ten years. The population growth is expected to continue, and in 2030, the
county is estimated to have more than 2.6 million inhabitants. (Länsstyrelsen Stockholm, 2017)
To cope with the increasing need for housing, the region needs a high housing construction
combined with increased mobility on the housing market. There is also a need for greater variety of
homes being built in order to meet many different needs on the market.
The housing construction in Stockholm fell drastically in the beginning of the 1990s as a result of
the financial crisis and the deregulation on the housing market. The low construction remained until
2013. However, in recent years, housing construction has increased sharply. From 2013 to 2016,
Stockholm had almost 62,000 housing construction starts, including reconstruction, in the county,
which can be compared to 29,000 starts in the previous four-year period. In 2016, the regional
yearly target of 16,000 new homes was exceeded for the first time, when 16,250 new homes were
built. However, despite increased housing construction, housing shortage increases as population
in the county grows at a faster pace than the housing construction. (Länsstyrelsen Stockholm, 2017)
Moreover, in recent years, many developers in Stockholm have focused on a niche with more
expensive housing. This has created an imbalance in the housing stock, resulting in difficulties for
households with lower income to enter the housing market. Some consider the more expensive
housing segment to be more or less saturated, while there is still a huge demand for cheaper housing
and rental apartments in Stockholm. (Dagens industri, 2017)
8
Moreover, the housing shortage in Stockholm can have a negative effect on the labour market.
Without housing for employees, the opportunities for companies to grow are limited. This may lead
to companies moving to other cities.
Furthermore, in order to meet the needs of a growing population in Stockholm, the public transport
system needs to be expanded. The subway system in Stockholm is under development. 19 km of
extended tracks, together with eleven new stations, will provide increased accessibility within the
Stockholm region. The development will create conditions for increased housing development, and
by agreements, the municipalities have undertaken to construct approximately 78,000 homes until
2030, within the metro area. Moreover, an expended and capacity-efficient railroad will increase
the opportunities for commuting in the county, which could facilitate housing development in new
areas.
Moreover, according to Newsec (2017), the demand for new housing has grown sharply in recent
years, partly due to increased immigration and urbanization. However, signs of satiation are now
shown. As shortage of labour and land leads to rising construction costs, and as interest rates begin
to rise, housing construction is expected to level out during 2018. Moreover, the longstanding trend
of rising housing prices in Stockholm has recently turned, meanwhile a number of regulations and
tax rises concerning the property sector are being discussed. These rules and regulations are causing
concerns, and they are likely to impact the housing and real estate market. Rules about households’
indebtedness leads to sharper lending regulations, which will increase the cost of mortgage.
(Newsec, 2017)
9
3 Literature review
This chapter presents a summary of existing literature on real estate cycles followed by housing
supply and demand determinants. Moreover, the chapter includes the current knowledge together
with theoretical and methodological contributions to the subject area.
3.1 Business and real estate cycles
Burns and Mitchell (1946) examined cyclical fluctuations of the market economic system. In their
book “Measuring business cycles” they provide the following definition of business cycles:
“Business cycles are a type of fluctuation found in the aggregate economic activity of nations
that organize their work mainly in business enterprises: a cycle consists of expansions
occurring at about the same time in many economic activities, followed by similarly general
recessions, contractions, and revivals which merge into the expansion phase of the next cycle;
in duration, business cycles vary from more than one year to ten or twelve years; they are not
divisible into shorter cycles of similar characteristics with amplitudes approximating their
own.” (Burns and Mitchell, 1946, p. 3)
One of the main reasons for studying business cycles is to identify and understand the unstable
nature of the market mechanisms. This also applies to real estate cycle analysis. (Reed et. al., 2010)
Wheaton (1999) defines a real estate cycle as follows:
“… a real estate cycle involves repeated oscillations of a market, as it continually overshoots
and then undershoots its own steady state … real estate cycles are defined as some degree of
instability in the market whereby a single economic shock leads the market to oscillate around
its steady state for some number of iterations” (Wheaton, 1999)
Fanning (2005) investigated the fundamental causes of business and real estate cycles. He also
studied the interaction of the national business cycle and local real estate cycles, and how the market
adjusts to changes in supply and demand.
10
Fanning (2005) categorises the real estate cycle into four phases:
Expansion – A period with increasing occupancy and absorption rates, rising prices and
eventually an upturn in construction activity.
Slowdown, peak and downturn – During this period, absorption rates start to decline, but are
still positive, and construction activity begins to fall. The already high real estate prices
continue to rise, while occupancy remains at a high level. When the peak of the cycle is
reached, absorption rates and construction activities start to decline.
Contraction – A period when occupancy and absorption rates decline, and prices for real
estate fall. This situation eventually leads to a decline in construction activity.
Slowed contraction, trough, and upturn – During this period, the downward trend of
construction activity reverses, and absorption and occupancy rates stabilize. After reaching
the lowest point of the cycle, the real estate market commences a moderate upturn, leading to
the next expansion phase.
Furthermore, two different types of real estate cycles are often distinguished; the physical cycle and
the financial cycle. The physical cycle describes fluctuations in construction activity, which is
affected by supply and demand, while the financial cycle, describes fluctuations in real estate prices,
which is affected by capital flows. Both cycles are related to each other, as well as to other
fluctuations, e.g. in vacancy levels and rent levels. (Rics, 2015)
Long term real estate cycles are primarily affected by the national changes in population and
income. A growing population and a rising income lead to an increase in demand for real estate. In
the short term, however, the availability of credit or capital, and the level of interest rates are the
main determinants of real estate demand. Changes in interest rates affect the price paid for capital.
Since real estate usually requires a large amount of debt financing, the demand for real estate is
highly affected by changes in interest rates. When interest rates decline, the demand for real estate
increases. (Fanning, 2005)
Previous research indicates that construction lags in relation to demand changes for space is the
primarily reason for cyclical behaviour on the real estate market, identified in the long run. Changes
in demand is mainly driven by fluctuations in business activity. Barras (1994) shows how each step
in the real estate cycle can be linked to the parallel business cycle, and how the fundamental
structure of the real estate market, with building booms and recessions, are affected by real and
money economy in the business cycle. This relationship is illustrated in figure 2.
11
Figure 2. An illustration of how each step in the real estate cycle can be linked to the real and money economy in the parallel business cycle. (Barras, 1994)
Moreover, time lags in the real estate cycle occur as the market attempts to adjust to changes in
supply and demand. When there is an excess in demand, prices and rents begin to rise before the
construction of new real estate starts to accelerate. Excess in supply also affect the real estate
market. After a period with rising prices and rents, construction starts to accelerate, leading to excess
space. Even when the excess in supply is recognized, buildings that are currently under construction
still have to be completed. The construction lag, resulting from the time it takes to complete the
building process, leads to an even larger excess in supply at the completion of the new buildings,
since more stock has been added to the market. (Fanning, 2005) Figure 3 shows how the real estate
cycle is lagged to the business cycle.
12
Figure 3. An illustration of the lag between the real estate cycle and the business cycle. (Fanning, 2005)
Moreover, Wheaton (1999) constructed a stock-flow model to study the office rental market and its
cyclical behaviour of price development. He suggested that the supply of real estate is almost
inelastic in the short term, meaning that the price for real estate is stabilized as demand meets supply
in a market with constant supply.
However, this approximation is not applicable in the long run, where rent affects construction, and
construction, in a longer perspective, affects supply. Consequently, the price elasticity of supply
can be interpreted as the price elasticity of construction.
Moreover, the stock-flow model shows that the main factor causing cycles is the supply lag, which
is caused by the time lag in construction of buildings. Wheaton (1999) questioned the idea of
consistent volatility in the real estate cycle. He argues that longer construction lags lead to larger
cycles. Differences between time lags of construction in different real estate sectors lead to
differences in the risk of overproduction, meaning that fluctuations depend on the type of real estate,
where retail and office properties are more prone to unpredictable changes in real estate cycles
compared to residential properties.
Office and hotel markets are at greater risk of overproduction due to longer construction lags and
higher uncertainty on the demand side. Residential market, on the other hand, experiences less
uncertainty on the demand side, meaning that the risk of overproduction is lower. The construction
lags for industrial real estate are shorter, which consequently means that the risk of overproduction
is lower in this sector.
13
Furthermore, Wheaton (1999) argues that a high elasticity of demand reduces real estate cycles,
while a high elasticity of supply is a precondition for cycles. The larger the differences between
elasticity in supply and demand, the larger are the cycles. Additionally, Wheaton also suggests that
myopic expectations leads to oscillations in the real estate cycle, while forward forecasting and
rational expectations leads to smooth adjustment and stability. However, there are conflicts between
rationality based and behaviour based theories.
3.2 Behaviour based theories
Brunes (2006) investigated three behavioural aspects; myopic behaviour, herd behaviour and
overconfidence, and their influence on decisions made by investors in commercial buildings, up to
the stage of the start of construction. A statistical part of the study was based on Tobin’s q-theory
of investments, which states that investment decisions are based on the ratio of the price of an
existing recently built property to the cost of production, also called Tobin’s q.
Brunes constructed a regression model to analyse construction volume and Tobin’s q with different
time lags. Strong correlation between the volume of investment in a specific year, and Tobin's q in
that specific year, could indicate myopic behaviour. Brunes found that the explanatory power of a
model with a three-year lag on Tobin’s q was quite good. This could be interpreted as if investors
in commercial buildings are characterized by myopic behaviour since the lag between the decision
to build, and the completion of commercial buildings, is about three years on average.
However, Brunes hypothesis of herd behaviour and overconfidence among investors in commercial
buildings did not get as strong support in his study as his hypothesis about myopic behaviour.
(Brunes, 2006)
3.3 The q theory of housing investment
Tobin (1969) presented the idea that the “rate of investment – the speed at which investors wish to
increase the capital stock – should be related, if to anything, to q, the value of capital relative to its
replacement cost”. This idea is also known as the q theory, and the ratio, q, is known as Tobin’s q
or the q ratio. Applied to the housing market, Tobin’s q can be expressed as the market price of an
existing building divided by the total production cost for a similar building.
According to the q theory, it is profitable for housing suppliers to build when Tobin’s q is greater
than one (q>1), since this indicates excess in demand on the housing market. On the other hand,
14
when Tobin’s q is less than one (q<1), there is an excess in supply, and housing production should
stop. In this situation, it is cheaper to buy an existing house compared to a newly built one.
Several studies have found a clear statistical relationship between the size of Tobin’s q and the
construction activity, both over time, and in different regions.
Jud and Winkler (2003) applied the q theory to the housing market. Using quarterly data for the
United States, they present estimates of a q model of housing investment. Their results show a clear
statistical relationship between the current and lagged values of the q ratio and housing investment.
This suggests that housing developers seem to respond to the demands of housing consumers,
building more when the prices of existing homes are high compared to the prices of new homes.
Furthermore, Berg and Berger (2006) investigated the existence of a stable relationship between
Tobin’s q and the investment in owner-occupied houses, on the Swedish housing market between
1981 and 2003. Berg and Berger suggest that the major changes in economic policy in Sweden, in
the early 1990s, resulted in a more market driven demand for housing investment. Their results
indicate that there is a high degree of correlation between the housing investment and the q ratio for
the last period of the sample, 1993-2003. However, a stable long-run relationship could not be
detected between 1981 and 1992.
Moreover, Takala and Tuomala (1990) studied the housing investment in Finland between 1972
and 1987. Their findings suggest that the q ratio is a significant determinant for housing investment,
after 1980.
3.4 Housing bubbles
According to Glaeser et al. (2008), housing price volatility are sometimes too high to be explained
by fundamentals of the market. This suggests that over-optimism and irrational behaviour among
buyers on the housing market could cause housing bubbles. Stiglitz (1990) defined a bubble as
“If the reason that the price is high today is only because investors believe that the selling
price will be high tomorrow – when “fundamental factors” does not seem to justify such a
price – then a bubble exists.”
Glaeser et al. (2008) argues that the volatility of housing prices differs across market with different
supply elasticity. By examining housing prices, new construction, and supply elasticity during
booms and busts, they showed that places with a more inelastic supply had much larger increases
in prices and smaller increases in new construction during booms compared to more elastic places.
15
However, places with more elastic supply might suffer from larger welfare losses, as a result of a
housing bubble, since there are more overbuilding in these places during a bubble.
3.5 The DiPasquale-Wheaton model
DiPasquale and Wheaton (1992), divided the real estate market into two interrelated markets: the
market for real estate space and the market for real estate assets. Moreover, they presented a model
that illustrates the important connection between these two markets, and how they are affected by
macroeconomic factors and financial markets. The model demonstrates the effects, resulting from
various exogenous shocks on rents, prices, construction and the stock of real estate. The model is
presented in a four-quadrant diagram, which illustrates how changes in both the nation’s macro
economy and its financial markets results in new equilibrium in the asset and property markets.
(See figure 4)
Figure 4. The DiPasquale-Wheaton model, illustrating the connection between the property market and the asset market. (DiPasquale and Wheaton, 1992)
Moreover, DiPasquale and Wheaton (1992) argued that an increase in employment, the number of
households, or household income would increase the demand for space, leading to rising rents and
housing prices, which in turn leads to an increase in construction. Furthermore, the prices of real
estate assets, in relation to its cost of construction, affect the supply of real estate. In the long run,
16
market prices and construction cost will be equal, while in the short term, these two may differ
significantly because of the lags inherent in the construction process. (DiPasquale and Wheaton,
1992)
17
4 Methodology and research design
The following chapter aims to describe and justify the overall methodological approach for
investigating and answer the research question; “Can housing construction in Stockholm be
explained by fundamental economic factors, and if, to what extent?” A statistical analysis
framework is presented as well as a description of how the thesis is conducted. Finally, the
regression method is described together with a description of statistical tests performed on the
regression model.
Due to the nature of the research problem, a deductive approach was applied. A deductive approach
begins with the development of hypotheses based on existing theory, after which the hypotheses
are tested by applying relevant methods. This either leads to the confirmation or rejection of the
hypotheses. The general steps in studies using a deductive approach is presented in figure 5.
(Saunders, et.al., 2016)
Figure 5. The steps in studies using a deductive approach
A quantitative method was used to answer the research question in this thesis. First, a literature
review was conducted, after which the quantitative analysis was performed as a regression analysis.
4.1 Literature review as a theoretical basis for the research
First, to increase knowledge, and to achieve better understanding of the research subject, a literature
review on previous studies on housing supply and demand determinants, as well as real estate
cycles, was conducted. Mostly academic papers from economic journals were used in the literature
review, but also government reports, and market reports published by well established companies,
have been used to increase knowledge and to obtain information about Stockholm County and its
housing market. The literature review serves as a base for choosing the variables used in the
Theory Hypothesis Observation/Test Confirmation/Rejection
18
regression analysis, as well as it investigates existing theory, which the hypotheses, formed in this
thesis, are based on. Moreover, the literature review also serves as a complement to the regression
model, giving the thesis a broader perspective. The literature review is presented in chapter 3.
4.2 Quantitative method
A quantitative analysis was carried out as a regression analysis, where data on housing construction
in Stockholm County from 1992 to 2016 was analysed as the dependent variable. The regression
analysis covers fundamental factors, serving as independent variables in the regression model. The
following eight independent variables were examined: Repo rate, inflation, GDP, population,
income, unemployment, housing prices, and construction cost.
The layout of the quantitative study in this thesis consists of the following steps: Firstly, the
collection and transformation of data. Secondly, the modelling of a well fitted and reliable model.
And finally, the interpretation and analysis of the results.
4.3 Regression analysis
Regression analysis is a collection of statistical techniques for investigating the relationship
between variables. A regression model aims at modelling the effect of a given set of independent
variables 𝑥", 𝑥$, … , 𝑥& on a variable 𝑦 of primary interest, called the dependent variable. (Fahrmeir
et al., 2013)
4.3.1 Linearregressionmodels
A commonly used type of regression model is the linear regression model, in which the dependent
variable, 𝑦, is a linear function of the unknown regression coefficients 𝛽), 𝛽", … , 𝛽&, and the
independent variables 𝑥", 𝑥$, … , 𝑥&. The general multiple linear regression model can be written as
𝑦 = 𝛽) + 𝛽"𝑥" + 𝛽$𝑥$ + ⋯+ 𝛽&𝑥& + 𝜀
Where 𝛽) is the intercept, and 𝜀 is the error term which contains unobserved factors, that is, factors
other than the independent variables in the model, that affect the dependent variable. (Wooldridge,
2013)
Linear regression models are commonly used in empirical studies that investigate complex and
unknown functional relationship between a dependent variable and some independent variables.
(Myers et al., 2012) This thesis uses multiple linear regression models on time series data to analyse
what fundamental economic factors that affect housing construction in Stockholm. Moreover, a
19
technique called Ordinary Least Squares (OLS) is used to estimate the coefficients in the regression
model. The coefficients are obtained by minimizing the sum of the squared differences between a
linear combination of the independent variables and the dependent variable.
4.3.2 Distributedlagmodels
Previous studies have shown that it takes time for an economy to fully react to changes in supply.
The full reaction may appear after several time periods. When an independent variable is expected
to affect the dependent variable after a time delay, the independent variable must be lagged in the
regression model. This can be done in a distributed lag model, where one or more independent
variables can affect the dependent variable with a time lag. (Wooldridge, 2013)
If 𝑥" is expected to affect 𝑦 with a lag of one period, while 𝑥$ is expected to have an instant impact
on 𝑦, the following simple lag model can be used
𝑦. = 𝛽) + 𝛽"𝑥"./" + 𝛽$𝑥$. + 𝜀.
Moreover, if the regression model predicts the current value of a dependent variable 𝑦, based on
both the current value, and different lagged values, of an independent variable 𝑥, the following
distributed lag model is applicable
𝑦. = 𝛽)𝑥. + 𝛽"𝑥./" + 𝛽$𝑥./$ + ⋯+ 𝛽0𝑥./0 + 𝜀
However, lagged variables are associated with some problems, such as multicollinearity and an
increased degree of freedom in the model. These problems can be checked using various tests
presented in chapter 4.4. (Studenmund, 2014)
4.3.3 Timetrend
It is common for economic time series to have a common tendency to grow over time. In order to
draw causal inference using time series data, it is important to identify these time trends. Ignoring
trending behaviour among variables in the regression model can lead to false conclusions that
changes in one variable are caused by changes in another variable. However, a time trend variable
can be included in the regression analysis to eliminate problems with spurious regression caused by
trending variables. (Wooldridge, 2013)
Moreover, If the dependent and independent variable have different types of trends, e.g. one variable
is trending upward and one downward, the inclusion of a time trend can increase the significance
of the independent variable. (Wooldridge, 2013)
20
To account for the trending behaviour of the variables included in the regression analysis in this
thesis, a time trend has been added to the regression model.
4.4 Statistical tests of the data and models
4.4.1 Stationarity
A stationary time series means that the variable has a constant mean and variance over time. In
order to get a valuable result from the regression analysis, all time series in the model need to be
stationary. However, it is common for time series to be non-stationary, and thus, show a random
walk. The Dickey-Fuller test can be used to check for stationarity among all variables in the
regression model. To deal with problems of non-stationarity, the time series can be transformed in
several ways, e.g. through logarithmic transformation or first differentiation.
4.4.2 Correlation
Correlation is a statistic measurement that describes the degree of relationship between two
variables. Correlation is described by a value between -1 and 1, where 1 describes perfect
correlation, meaning that the variables follow each other to 100% and -1 describes perfect negative
correlation, meaning that the variables move in opposite directions. A correlation coefficient of zero
implies that there is no relationship between the variables, and thus, they move totally independent
of each other.
4.4.3 Multicollinearity
Multicollinearity is a phenomenon in which there is a high correlation between two or more
independent variables in a multiple linear regression model. Since multicollinearity usually leads to
less reliable p-values for the independent variables together with wider confidence intervals, it can
cause difficulties in distinguishing the individual effects of the correlated variables in the regression
model.
To identify multicollinearity in a model, a variance inflation factor (VIF) analysis can be conducted.
The variance inflation factor measures to what extent a variable is contributing to the standard error
in the regression model, where a large VIF value indicates that issues with multicollinearity exist
in the model. Usually, some degree of multicollinearity exists, and it is therefore important to
identify whether it constitutes a problem in the regression model. After identifying variables with
multicollinearity issues, correlated variables can be combined or removed to solve the problem.
Furthermore, multicollinearity caused by common trends can be eliminated by transforming the
time series using the first differences or the logarithms. (Studenmund, 2014)
21
4.4.4 Autocorrelation
Autocorrelation is the correlation between a given time series and a lagged version of the same time
series. In other words, autocorrelation describes the degree of similarity between observations of
the same variable over a period of time. If the observations are autocorrelated it means that future
observations are affected by prior values, and, consequently, the observations are not random.
Furthermore, autocorrelation occurs most frequently in time series data sets. Since autocorrelation
can cause bias it is important to test for autocorrelation among all variables in a regression model.
The most commonly used test for autocorrelation is the Durbin-Watson d Test. Moreover, Durbin’s
alternative test for autocorrelation can be used as a more general test to determining whether there
is any evidence of autocorrelation. This test considers more than one lag, and it is therefore suitable
for regression models that includes several time lags. (Studenmund, 2014)
4.4.5 Heteroscedasticity
Heteroscedasticity describes a situation in which the size of the error term differs across values of
an independent variable. When the variance for the distribution of the error term is non-constant,
the OLS regression tends to generate inaccurate estimates of the standard error of the coefficients,
meaning that the standard errors are bias. Since the standard error plays a central role when
conducting significance tests, bias can lead to incorrect conclusions about the significance of the
coefficients in the regression model. The opposite of heteroscedasticity is homoscedasticity, which
means that the error term has a constant variance. Figure 6 illustrates the difference between homo-
and heteroscedastic variances. (Studenmund, 2014)
Moreover, the White test can be used to test for heteroscedasticity. Problems with heteroscedasticity
can be solved by applying logarithmic transformation to the heteroskedastic variable. (Studenmund,
2014)
Figure 6. Illustration of homo- and heteroscedastic variances. a) funnel-shaped heteroscedastic variance, errors. b) homoscedastic variance, errors. (Fahrmeir et al., 2013)
22
5 Data collection and hypotheses
This chapter describes the data collection process, the data sources used and the transformation
of the data. Furthermore, the dependent, and all the independent variables that will be included in
the regression analysis are presented, together with their theoretical impact on housing
construction as well as a derived hypothesis for each independent variable.
Berg and Berger (2006) argued that the changes in tax and housing policy in Sweden in the end of
the 1980s and the beginning of the 1990s led to a more market driven housing demand. This means
that determinants for housing investments might differ between the old and the new policy regime.
According to Berg and Berger (2006) the full effect of the policy changes could be seen around
1992 and 1993. This forms the basis for the choice of time span for the data included in the
regression in this thesis. All variables used in the regression analysis consist of time series data
compiled on a quarterly basis for the period 1992-2016.
The data used in the regression analysis were collected from two different sources. Most data were
retrieved from Statistics Sweden (SCB), but also the central bank of Sweden (Riksbanken) was used
as a source for the data collection. All variables are presented in table 1, along with their data
sources. Furthermore, nominal values are given for all variables containing monetary data. This
means that the effect of the inflation on these variables is the same over time.
During the data collection, problems arose with finding some of the data on a quarterly basis.
Population, income, and unemployment were only found on a yearly basis. Data on the annual
percentage average of unemployment in Sweden were transformed into quarterly data by applying
the same unemployment rate for quarter 1-4 for each year.
The yearly data on population and income were transformed into quarterly data by dividing the
change in population and income, from one year to the next, by four. This transformation method
is motivated by the fact that these variables are less volatile and the growth of these variables are
approximately continuous. Furthermore, the consumer price index (CPI) was used as a
measurement of inflation. Data on CPI was only found monthly. The monthly data was transformed
into quarterly data by collecting the consumer price index corresponding to the first month in each
quarter. Moreover, the income and unemployment data refer to Sweden, and not specifically to
Stockholm County.
23
Table 1. Regression variables and their data sources
Variable Source Housing construction SCB Repo rate Riksbanken Inflation SCB GDP SCB Population SCB Income SCB Unemployment SCB Housing prices SCB Construction cost SCB
5.1 Variables
5.1.1 Thedependentvariable
Housing construction
The dependent variable that is aimed to be explained in the regression model is the housing
construction in Stockholm County. The housing construction data collected comprises the number
of housing starts in newly constructed multi-dwelling buildings in Stockholm County during the
period 1992-2016. The data have been retrieved from Statistics Sweden.
5.1.2 Independentvariablesandhypotheses
For each independent variable, a hypothesis is formulated, based on previous studies presented in
the literature review. All hypotheses denote the effect on the dependant variable, housing
construction, resulting from a positive change in the independent variable, when all other variables
are held constant.
The null hypothesis states that the independent variable does not have an effect on the dependent
variable, which can be written as
H0: 𝛽1 = 0, where 𝑗 = 1, 2, 3, … , 9
All hypotheses are summarized in table 2.
24
Repo rate
The repo rate is the overnight interest rate offered to banks by the central bank of Sweden,
Riksbanken, and it is the primary instrument for adjusting inflation and economic growth. The
central bank can reduce the repo rate to stimulate consumption in the country, and thereby increase
inflation, or vice versa.
The repo rate was first introduced in Sweden in 1994. Before that, from December 1985 until May
1994, the marginal rate was the central bank of Sweden's policy rate. For the first nine data
observations in the regression model in this thesis (1992 Q1 – 1994 Q1), the marginal rate is used
as a proxy for the repo rate.
Furthermore, the repo rate in Sweden is forecasted to rise at a slow pace, but it is expected to remain
at a low level over a foreseeable future. (Riksbanken, 2017) Mortgage rate follows the changes in
the repo rate, and since low interest rate leads to decreased cost of mortgage capital for households,
a low interest rate environment is expected to increase housing prices, which in turn is expected to
lead to increased housing construction.
H1: A positive change in repo rate yields a decrease in housing construction
Inflation
Inflation is a sustained rise in the general level of prices of goods and services in an economy. With
a rising price level, fewer goods and services can be bought with a unit of currency, meaning that
inflation leads to a reduction in the purchasing power. A common measure of inflation is the
inflation rate, which is the annual percentages change in a general price index. (Blanchard, 2000)
The central bank of Sweden uses changes in the consumer price index, CPI, as a measurement of
the inflation rate in Sweden. (Riksbanken, 2018)
Several studies have examined if investment in real estate could hedge against expected and
unexpected inflation. The general result from these studies show that real estate provides a partial
hedge against some components of inflation (Chaney and Hoesli, 2010). Moreover, Schwab (1982)
investigated the relationship between inflation expectations and the demand for housing. According
to Schwab, some consumer uses housing as a hedge against expected inflation. This means that a
positive change in the inflation rate is expected to decrease the demand for housing, which in turn
leads to a decrease in housing construction.
H2: A positive change in inflation yields a decrease in housing construction
25
GDP
The Gross Domestic Product, or GDP, is the value of all the goods and services produced in a
country during a specific time period, usually one year. GDP is a way to measure the economic
activity in a country and it is commonly used as an indicator of a country’s economic performance.
GDP serves as a measurement of the size of a country’s economy, while GDP growth describes the
performance of the country from year to year.
Demand and supply are fundamental factors that lead to the construction of GDP, which plays an
important role in the business cycle. A growing economy usually leads to increase in demand for
goods and services as well as an increase in supply of these goods and services. Based on this, a
positive GDP growth can be expected to increase the demand for housing, and thus also increase
housing construction.
Moreover, Barras (1994) showed how the real estate cycle can be linked to the parallel business
cycle, and how an economic upturn leads to increased property demand, which in turn leads to
supply shortage, after which rents rise, eventually leading to a building boom.
Moreover, DiPasquale and Wheaton (1992) argued that economic growth increases all equilibrium
variables, i.e. rent, price, construction, and stock, in the real estate market.
H3: A positive change in GDP yields an increase in housing construction
Population
Fanning (2005) argued that long term real estate cycles are partly affected by the national changes
in population, where a growing population leads to an increase in demand for real estate.
Moreover, according to DiPasquale and Wheaton (1992), the demand for space depends, among
other factors, on the number of households, which is closely correlated to the population. All else
equal, an increase in the number of households will increase the demand for space, hence the
housing construction increases, which eventually leads to an increase in the housing supply.
H4: A positive change in population growth yields an increase in housing construction
Income
Fanning (2005) suggested that long term real estate cycles are partly affected by the national
changes in income, where a rising income lead to an increase in demand for real estate.
Moreover, DiPasquale and Wheaton (1992) argued that the household demand for space partly
depends on household income, since the annual payments that a household can afford to pay are
26
primary determined by its level of income. An increase in household income is therefore expected
to increase housing demand, leading to an increase in housing construction.
H5: A positive change in income yields an increase in housing construction
Unemployment
Unemployment is the total number of people who are not employed but are actively searching for
employment. Unemployment is usually measured as the unemployment rate, which is the number
of people unemployed divided by all people currently in the labour force. (Blanchard, 2010)
DiPasquale and Wheaton (1992) argued that, among other variables, employment affects the
demand for space. An increase in employment increases the demand for space, leading to an
increase in rents. This in turn, leads to higher asset prices which generate an upturn in construction.
H6: A positive change in unemployment yields a decrease in housing construction
Housing prices
Because of difficulties in finding data on housing prices for multi-dwelling buildings from 1992 to
2016, a real estate price index, which refers to one- and two-family houses is used as a proxy for
housing prices for multi-dwelling buildings. A correlation test, and a comparison between the real
estate price index and data on housing prices for multi-dwelling buildings during the period 2005-
2016 (available from Statistic Sweden) showed a high degree of correlation and similarities between
these two data sets.
DiPasquale and Wheaton (1992) argue that housing prices largely depends on supply and demand
for housing. All else equal, an increase in demand for housing will raise housing prices, while an
increase in housing supply will lower the prices.
Mayer and Somerville (2000) presented an empirical model of housing supply, which describes
new housing construction as a function of changes in current and lagged housing prices. Their
findings suggest that a 10% increase in real house prices leads to an 0.8% rise in housing supply.
H7: A positive change in housing prices yields an increase in housing construction
Construction cost
To capture the effect of a change in construction cost in the regression model, an index called
building price index (BPI) will be used. The building price index measures the development of
prices for constructing new residential buildings, adjusted for quality and regional differences. The
27
building price index consists of construction cost and connection fees, for example, electricity and
district heating, as well as administrative costs, design costs, interest expenses, and value-added tax.
The building price index used in this thesis is based on prices for multi-dwelling buildings.
According to the q theory, it is profitable for housing suppliers to build when Tobin’s q is greater
than one, which means that the market price of an existing building is higher than the total
production cost for a similar building. The q theory of housing investment implies that, all else
equal, an increase in the construction cost would decrease construction.
H8: A positive change in construction cost yields a decrease in housing construction
Time trend
To account for the trending behaviour of the variables in the regression model, a time trend has
been added to the regression model. The independent variable shows a growing trend over time.
This suggests that an increase in the time trend variable yields an increase in housing construction.
H9: A positive change in time yields an increase in housing construction
Table 2. Summary of hypotheses
Variable Prediction
H1 Repo rate - H2 Inflation - H3 GDP + H4 Population + H5 Income + H6 Unemployment - H7 Housing prices + H8 Construction cost - H9 Time trend +
28
6 Results
This chapter presents the results from the regression analysis. The regression outcomes will be
presented followed by a summary of the hypothesis testing.
Figure 7 illustrates the changes in the dependent variable in the regression model. The figure shows
the quarterly number of housing starts in newly constructed multi-dwelling buildings in Stockholm
County between 1992 and 2016.
Figure 7. Quarterly number of housing starts in newly constructed multi-dwelling buildings in Stockholm County, 1992 Q1 - 2016 Q4
6.1 Regression outcomes
Table 3 presents a summary of the regression output. The full regression output is presented in
Appendix I. All variables in the regression model have been log transformed and/or differentiated
in order to retrieve stationarity among all time series. Tests for stationarity, using the Dickey-Fuller
test have been conducted. The test results indicated stationarity among all variables, except
population, which had a test statistic-value that was slightly higher than the 5% critical value.
Moreover, to retrieve statistically significant estimates, some variables have been lagged up to seven
periods, where one period equals one quarter. A correlation test has been conducted for all variables.
-
1 000
2 000
3 000
4 000
5 000
6 000
7 000
8 000
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
Housingstarts
Quantity
Year
29
The test did not show any problems with correlation among any of the variables. The correlation
test and the Dickey-Fuller tests are presented in Appendix II.
Furthermore, the regression model has been tested for multicollinearity, autocorrelation, and
heteroscedasticity, using the variance inflation factor (VIF), Durbin’s alternative test, and White’s
test, respectively. No problems with multicollinearity, autocorrelation or heteroscedasticity were
encountered. However, since most economic variables are heteroscedastic by nature, the commando
vce(robust) has been applied to the regression model. This commando improves models that might
suffer from heteroscedasticity, by making the standard errors robust. The commando improves the
standard errors, but it does not affect the 𝛽 coefficients in the model. For detailed test statistics, see
Appendix II.
Moreover, the model has an R-squared value of 0.73, which indicates that the model can explain
73% of the variation in housing construction in Stockholm County between 1992 and 2016.
Table 3. Regression results
Linear regression Number of obs 91 R-squared 0.7331 Robust Housing construction Lag β Coef. Std. Err. t P > ǀ t ǀ
Repo rate Lag 7 -0.2792 0.1234 -2.26 0.026** Inflation Lag 7 0.1521 0.0697 2.18 0.032** GDP Lag 0 4.0898 1.0599 3.86 0.000*** Population Lag 1 -211.4862 51.9763 -4.07 0.000*** Unemployment Lag 6 -1.1323 0.5182 -2.19 0.032** Housing prices Lag 2 6.4807 1.9787 3.28 0.002*** Construction cost Lag 3 -0.9505 0.5484 -1.73 0.087* Time trend - 0.0296 0.0022 13.41 0.000*** _cons - 5.8683 0.1623 36.15 0.000***
Notes: *, **, *** denote significance at 90%, 95%, and 99%, respectively.
The model presented in table 3 aims to explain the housing construction in Stockholm County
between 1992 and 2016. P-values and 𝛽 coefficients are presented for all independent variables
included in the model. Income was not statistically significant, and the variable was therefore
excluded from the model.
30
The p-values are used to determine statistical significance in the hypothesis testing. If the p-value
is less than 0.05, the null hypothesis can be rejected, meaning that the corresponding independent
variable has a significant effect on the dependent variable to a 95% significance level.
The 𝛽 coefficient describes to what extent a change in a variable affect housing construction in
Stockholm County. A positive value of the 𝛽 coefficient means that a positive change in the
corresponding independent variable yields an increase in housing construction, while a negative
value of the 𝛽 coefficient means that a positive change in the independent variable yields a decrease
in housing construction.
Several variables were expected to have a lagged impact on housing construction. The model was
constructed using different periods of lag on the independent variables.
The final model includes Repo Rate lagged seven periods, Housing Prices lagged two periods,
Construction Cost lagged three periods, Population lagged one period, GDP without lag,
Unemployment lagged six periods, and Inflation lagged seven periods. Also, a time trend, capturing
trending behaviour of the model variables, was included in the regression model.
Furthermore, the dependent variable, Housing Construction, has been transformed using the natural
logarithm. Moreover, in order to obtain stationarity, the Repo Rate, Housing Prices, Construction
Cost, Population, and Unemployment have been transformed using the change in the natural
logarithm, which can be interpreted as the growth rate of the variables. GDP has been transformed
using the nature logarithm, and Inflation has been first differentiated.
31
6.2 Hypothesis testing
The hypothesis testing is presented in table 4, which includes a summary of the hypotheses together
with the outcome from the regression model. Six out of nine hypotheses were correctly predicted
to a significance level of 0.1 or above.
Table 4. Summary of the tested hypotheses with prediction and outcome
Variable Prediction Outcome Significance
level
H1 Repo rate - - ** H2 Inflation - + H3 GDP + + *** H4 Population + - H5 Income + omitted H6 Unemployment - - ** H7 Housing prices + + *** H8 Construction cost - - * H9 Time trend + + ***
Notes: *, **, *** correctly predicted and significant at 90%, 95%, and 99%, respectively.
Since the income variable is not significant at any acceptable level, there is not enough evidence to
reject the null hypothesis, which states that there is no linear relationship between income and
housing construction in Stockholm County. Therefore, the income variable is excluded from the
model. Moreover, the hypotheses for Repo Rate, GDP, Unemployment, Housing Prices,
Construction Cost, and the time trend were correctly predicted to a significant level of 0.1 or above,
while the hypotheses for inflation and population was wrong.
32
7 Analysis
This chapter presents an analysis and discussion regarding the results and hypothesis testing
presented in the previous chapter. All variables will be analysed and the discussions will
reconnect to the literature review. Moreover, reliability and validity, as well as the limitations of
the study, will be discussed in this chapter.
7.1 Analysis of variables
Six out of nine variables, examined in the regression analysis, match the predicted outcomes in the
regression, as well as theory from previous studies presented in the literature review. However,
there are some interesting and unexpected results that will be further discussed in this chapter.
Repo rate
A positive change in repo rate was expected to yield a decrease in housing construction. The
hypothesis testing shows that the null hypothesis can be rejected to a 95% significance level. Interest
rates, which are closely related to the repo rate, plays an important role for both the supply and
demand side of the housing market. If the interest rates decrease, it becomes cheaper, and thus more
profitable, for construction companies to build, which leads to an increase in supply. At the same
time, lower interest rates lead to decreased cost of mortgage capital for households, which in turn
leads to an increase in housing demand, as well as an increase in housing prices, which further
enhance the willingness for housing developers to increase housing construction.
Inflation
A positive change in inflation was expected to yield a decrease in housing construction. However,
the regression result showed a positive 𝛽 coefficient for the inflation variable. The result in the
regression was unexpected since literature describes investment in real estate as a way to hedge
against inflation. However, the result might be explained by the characteristics of booms and
recessions. Booms are associated with periods of high inflation. Moreover, according to Barras
(1994), economic booms are linked to building booms which in turn leads to increased supply. This
linkage might explain the positive 𝛽 coefficient for the inflation variable in the regression outcome.
33
GDP
A positive change in GDP was expected to yield an increase in housing construction. The hypothesis
testing shows that the null hypothesis can be rejected to a 99% significance level.
GDP growth describes the economic performance of the country. Moreover, supply and demand
serves as important factors in the construction of GDP. An increase in supply and demand for goods
and services can be expected to include an increase in the supply and demand for housing. During
times when the economy is flourishing, people are better off, and can therefore afford more housing,
which increase the demand for housing, leading to increased housing construction.
Population
A positive change in population was expected to yield an increase in housing construction.
However, the regression result showed a negative 𝛽 coefficient for the population variable. This
was a surprising and unexpected result. Literature suggest that the demand for space partly depends
on the number of households, where an increase in the number of households would increase
housing construction.
Several problems where encountered during the data collection and the data transformation process
regarding the data on population in Stockholm County. Firstly, the data was only found on a yearly
basis, which means that it had to be transformed into quarterly data. The transformation method
used, where the change in population from one year to the next was divided by four to obtain
quarterly data, generated a data set with a consistent population growth over each year. The actual
population growth is probably more volatile. If quarterly data had been available, the result from
the regression model might have been different and more accurate.
Moreover, the Dickey-Fuller test for stationarity showed a result indicating that the time series on
population growth might experience problems with non-stationarity. The Dickey-Fuller test showed
a test statistic-value slightly higher than the 5% critical value. The problem with non-stationarity
might depend on the consistent population growth, resulting from the data transformation.
However, although this result was unexpected, the underlying reason for this result could also arise
from several characteristics of the population growth. A more thoroughly analysis could have been
conducted if the driving forces for the population growth was known. Housing construction is partly
driven by the demand for housing, not by the need for housing. If the population growth mainly
derives from an increase in the number of job opportunities in Stockholm, the demand for housing
might rise. However, if the population growth is caused by increased immigration or increased
childbirth in the region, the demand for housing may not rise to the same extent, due to a lower
ability to pay for housing within these groups. Furthermore, the age distribution among the
34
population in a region can affect the demand for housing as well. A population with a high density
of children might demand less housing since the population might consist of larger families living
together. While a population consisting of a high density of young adults might increase the demand
for housing. Moreover, if the population growth comes from an increase in childbirth in the region,
the increase in the demand for housing might not show until after several years, and the time lag of
up to three years, used in the regression model in this thesis, might not be enough to capture the
effect of population growth.
Income
A positive change in income was expected to yield an increase in housing construction. However,
the hypothesis could not be confirmed nor denied since the regression output showed an
insignificant result for the income variable. Previous studies have suggested that the household
demand for space partly depends on household income, since the annual payments that a household
can afford to pay for housing are primary determined by its level of income.
However, the insignificance of the income variable might be explained by inaccurate income data.
The data was only found on a yearly basis, and it was therefore transformed into quarterly data by
dividing the income growth from one year to the next by four, leading to a consistent income growth
over each year. The actual income growth might be more volatile than the income data used in this
regression, which might affect the result of the regression. However, in general, income is expected
to grow over time, and a decline in income is rare. Therefore, the dataset is probably not too
inaccurate. Moreover, the income data refers to Sweden, and not specifically to Stockholm County.
While the percentage income growth can be expected to be approximately the same all over Sweden,
this approximation might be one of the reasons to the insignificance of the income variable in the
regression.
Unemployment
A positive change in unemployment was expected to yield a decrease in housing construction. The
hypothesis testing shows that the null hypothesis can be rejected to a 95% significance level.
DiPasquale and Wheaton (1992) argued that an increase in employment increases the demand for
space, which in turn should lead to an increase in construction. A rise in employment rate indicates
a situation where the financial position of the population increases, which could be expected to
increase the demand for housing. Moreover, an employment is generally a requirement for an
individual to be granted a loan at a bank. Consequently, an increase in employment could be
expected to increase the number of people with the ability to buy a house or an apartment.
35
Housing prices
A positive change in housing prices was expected to yield an increase in housing construction. The
hypothesis testing shows that the null hypothesis can be rejected to a 99% significance level. This
result is in accordance with Mayer and Somerville (2000), who presented an empirical model of
housing supply, which describes new housing construction as a function of changes in current and
lagged housing prices. Their model suggests that an increase in real house prices leads to a rise in
housing supply.
Construction cost
A positive change in construction cost was expected to yield a decrease in housing construction.
The hypothesis testing shows that the null hypothesis can be rejected to a 90% significance level.
It is reasonable that, all else equal, a rise in construction cost decreases housing construction, since
a rise in construction cost lowers the profitability for the construction companies. Moreover, this is
in accordance with the q theory of housing investment, which states that it is profitable for housing
suppliers to build when the market price of an existing building is higher than the total production
cost for a similar building.
Time trend
A positive change in time was expected to yield an increase in housing construction. The hypothesis
testing shows that the null hypothesis can be rejected to a 99% significance level.
The time trend is statistically significant, and its 𝛽 coefficient implies an approximate 3% increase
in housing construction per year, on average. The R-squared value in the regression model presented
in this thesis is much higher compared to a similar model where the time trend has been excluded.
This might partly be explained by the fact that a time trend captures other factors that are trending
over time, outside the variables included in the regression model.
7.2 Limitations
This study explores housing construction through eight different variables. Since there are many
other factors that may affect housing construction, the result in this study might hence suffer from
variable selection bias. However, a thorough literature study has been conducted, which serves as a
base for the variable selection.
Moreover, some difficulties arose during the data collection process. Data on income and population
were only found on a yearly basis. The transformation of the data into quarterly data might cause
problem, since these time series are only approximations of the variables and not the actual values.
36
Furthermore, since the independent variables might affect housing construction with multiple time
lags, the regression model could possibly be improved by including multiple time lags for each
variable, meaning that each independent variable could affect the dependent variable with multiple
time lags.
Moreover, one question to consider when constructing a regression model is whether one or more
of the independent variables could be jointly determined with the dependent variable, usually
through an equilibrium mechanism. For example, the dependent variable in this thesis, housing
construction, can be explained by several variables. However, some of these variables might, in
turn, depend on housing construction. This simultaneity problem could have been solved by
applying a method for estimating a simultaneous equations model, which is a model in the form of
a set of linear simultaneous equations. However, applying the method for estimating simultaneous
equations models is beyond the scope of this thesis, and therefore housing construction is regarded
as the only dependent variable in the regression model.
37
8 Conclusion
The following chapter presents a summary of the main findings in this study. Furthermore, ideas
for further research within the research area are proposed.
8.1 Concluding remarks
The overall purpose of this thesis was to analyse what fundamental economic factors that affects
housing construction in Stockholm. Furthermore, a multiple linear regression model was presented,
which aims to describe to what extent fundamental economic factors affect housing construction in
Stockholm County.
Moreover, the R-squared of 0.73, in the model, indicates that 73% of the variation in housing
construction in Stockholm County can be explained by repo rate, inflation, GDP, population,
unemployment, housing prices, construction cost, and a time trend. However, the regression model
showed a low significance level for income, thus this variable was excluded from the model.
Furthermore, a time trend was included in the regression model. The time trend showed statistically
significance, and its β coefficient implies an approximate 3% increase in housing construction per
year, on average.
There are several factors, not included in the regression model, that might affect housing
construction. The municipal planning monopoly implies that municipalities serve as key actors for
housing development. Their decisions might not be based on fundamental economic factors.
Moreover, also other policy regulations in Sweden might influence the development on the housing
market, such as changes in taxes, and the introduction of government subsidies for housing
construction. These factors are not included in the model.
8.2 Suggestions for further research
Future studies on housing construction in Stockholm could focus on qualitative factors, such as the
impact of the Swedish planning system. Moreover, it would be interesting to examine the driving
forces, for municipalities, to increase housing construction. This could be done by studying how
municipalities work with the development of detail development plans within the municipalities.
38
9 References Barras, R. (1994), Property and the economic cycle: building cycles revisited, Journal of Property Research, Vol. 11 pp.183-97.
Berg, L. and Berger, T. (2006), The Q theory and the Swedish housing market – an empirical test, The Journal of Real Estate Finance and Economics.
Blanchard, O. (2010). Macroeconomics - A European Perspective. 1st ed. Harlow, England: Paerson Education Limited.
Boulhouwer P.J. and de Vries P., (2002), Housing production in the Netherlands: A growing misbalance between state and market. Working Paper, OTB Research Institute, Delft.
Boverket (2011), Vad bestämmer bostadsinvesteringarna?. [Online] Available at: https://www.boverket.se/globalassets/publikationer/dokument/2011/vad-bestammer-bostadsinvesteringarna.pdf [Accessed 8 January 2018]
Boverket (2014a), Låst läge på bostadsmarknaden. [Online] Available at: https://www.boverket.se/globalassets/publikationer/dokument/2014/marknadsrapport-maj-2014.pdf [Accessed 8 January 2018]
Boverket (2014b) Under miljonprogrammet byggdes en miljon bostäder. [Online] Available at: http://www.boverket.se/sv/samhallsplanering/stadsutveckling/miljonprogrammet/ [Accessed 8 January 2018]
Boverket (2014c) Vad är en detaljplan. [Online] Available at: http://www.boverket.se/sv/PBL-kunskapsbanken/planering/detaljplan/detaljplaneinstrumentet/vad-ar-detaljplan1/ [Accessed 8 January 2018]
Boverket (2015), Behov av bostadsbyggande - Teori och metod samt en analys av behovet av bostäder till 2025. [Online] Available at: https://www.boverket.se/globalassets/publikationer/dokument/2015/behov-av-bostadsbyggande.pdf [Accessed 8 January 2018]
Boverket (2016) Översiktsplanen. [Online] Available at: http://www.boverket.se/sv/PBL-kunskapsbanken/planering/oversiktsplan/oversiktsplanens-funktion/ [Accessed 8 January 2018]
Boverket (2017a), Läget på bostadsmarknaden i riket. [Online] Available at: https://www.boverket.se//sv/samhallsplanering/bostadsplanering/bostadsmarknaden/bostadsmarknadsenkaten-i-korthet/bostadsmarknaden-i-regioner-och-kommuner/riket/ [Accessed 8 January 2018]
Boverket (2017b). Boverkets indikatorer – Analys av utvecklingen på bygg- och bostadsmarknaden med byggprognos. [Online] Available at: https://www.boverket.se/globalassets/publikationer/dokument/2017/boverkets-indikatorer-maj-2017.pdf [Accessed 11 February 2018]
39
Brunes, F. (2006) Overbuilding in office markets: Are behavioural aspects important? Department of Infrastructure Building and Real Estate Economics, Royal Institute of Technology (KTH), Stockholm
Burns, A.F., Mitchell, W.C. (1946), Measuring Business Cycles, National Bureau of Economic Research, New York, NY,.
Chaney, A. and Hoesli, M. (2010). The interest rate sensitivity of real estate. Journal of Property Research, 27(1), pp.61-85.
Dagens industri (2017) Boverkets varning: För många som bygger dyra bostäder. [Online] Available at: https://www.di.se/nyheter/boverkets-varning-for-manga-som-bygger-dyra-bostader/ [Accessed 20 November 2017]
DiPasquale, D. & Wheaton, W.C., (1992). The markets for real estate assets and space: A conceptual framework. Journal of the American Real Estate and Urban Economics Association, vol. 20, pp. 181-197.
Emanuelsson, R. (2015). Utbudet av bostäder i Sverige. Sveriges riksbank.
Fahrmeir, L., Kneib, T., Lang, S. and Marx, B. (2013). Regression - Models, Methods and Applications. Springer-Verlag Berlin Heidelberg.
Fanning, S. F. (2005). Market Analysis for Real Estate - Concepts and Applications in Valuation and Highest and Best Use. Chicago: Appraisal Institute.
Glaeser E. & Gyourko J., 2002. The impact of building restrictions on housing affordability. Economic Policy Review,
Jud, G.D., Winkler, D.T. (2003), The Q theory of housing investment, Journal of Real Estate Finance and Economics, Vol. 27
Lind, Hans (2003), Ds 2003:6 Bostadsbyggandets hinderbana – en ESO-rapport om utvecklingen 1995-2001
Luger M.I. & Temkin K., 2000, Red tape and housing costs: How regulation affects new residential development. Center for Urban Policy Research, Rutgers University, New Jersey 2000.
Länsstyrelsen Stockholm (2017). Läget i länet – bostadsmarknaden i Stockholms län 2017. [Online] Available at: http://www.lansstyrelsen.se/Stockholm/SiteCollectionDocuments/Sv/publikationer/2017/R2017-15-Laget-i-lanet-bostadmarknaden-i-stockholms-lan.pdf [Accessed 11 February 2018]
Mayer, C. and Somerville, C. (2000). Land use regulation and new construction. Regional Science and Urban Economics, 30(6), pp.639-662.
Myers, Montgomery, Vining, and Robinson, ed., (2012). Linear Regression Models. In: Generalized Linear models, 2nd ed. John Wiley & Sons, Inc.
40
Newsec, (2017). Property Outlook. autumn 2017. Newsec. [Online] Available at: https://www.newsec.com/globalassets/documents/sverige/npo_autumn_2017.pdf [Accessed 8 January 2018]
Reed, Richard and Wu, Hao (2010), Understanding property cycles in a residential market, Property management
Rics (2015) Discover what a property cycle is, how long they last and how they can affect the property market. [Online] Available at: http://www.rics.org/se/knowledge/glossary/property-cycles/ [Accessed 11 February 2018]
Robert M. Schwab (1982). Inflation Expectations and the Demand for Housing. The American Economic Review, Vol. 72, No. 1, pp. 143-153
Saunders, M., Lewis, P. and Thornhill, A. (2016). Research methods for business students. Harlow: Pearson Education Limited.
SCB (2015), Bostadsbyggandet ökar kraftigt. [Online] Available at: https://www.scb.se/sv_/Hitta-statistik/Artiklar/Bostadsbyggandet-okar-kraftigt/ [Accessed 2 February 2018]
SFS 2010:900. Planning and Building Act. Available at: https://www.boverket.se/globalassets/publikationer/dokument/2016/legislation.pdf [Accessed 12 February 2018]
Stiglitz, J.E. (2000) Economics. W.W. Norton & Company, New York.
Studenmund, A. H., 2014. Using Econometrics a Practical Guide. 6th ed. Edinburgh: Pearson Education Limited.
Sveriges Byggindustrier (2015), Bostadsinvesteringar. [Online] Available at: https://www.sverigesbyggindustrier.se/bygginvesteringar/bostadsinvesteringar__6908 [Accessed 16 February 2018]
Sveriges Byggindustrier, (2016) Bostäder. [Online] Available at: https://www.sverigesbyggindustrier.se/statistik-byggmarknad/bostader__6904 [Accessed 11 February 2018]
Takala, K., and Tuomala, M. (1990). Housing Investment in Finland, Finish Economic Papers 3, 41-53.
Tobin, J. (1969). A General Equilibrium Approach to Monetary Theory, Journal of Political Economy 96, 718-740.
Wheaton, W.C. (1999), Real estate ‘cycles’: some fundamentals, Real Estate Economics, Vol. 27.
Wooldridge, J.M. (2013). Introductory econometrics: A modern approach (5thed.). Mason, OH: South-Western, Cengage Learning.
41
Appendices
Appendix I. Regression model
Variables Abbreviation
Housing construction Byggande
Repo rate REPO
Inflation INF
GDP GDP
Population POPG
Income INC
Unemployment UNEMP
Housing prices HOP
Construction cost BPI
Time trend quarter
Regression output
42
Appendix II. Statistical testing
Test for Stationarity
Reporate
Housing prices
Construction cost
43
Population
GDP
Unemployment
44
Inflation
Test for Correlation
Test for Multicollinearity
45
Test for Autocorrelation
Test for Heteroscedasticity
1
TRITA-ABE-MBT-184
www.kth.se